In the Explained series of blog posts, we break down complex technologies incorporated in our artificial intelligence, Aplysia. In this entry, we are exploring syntax and semantic analysis. We start with a broader overview and move on to specific features incorporated in our AI and show how they work with practical examples.
At HiJiffy, we are at the forefront of creating intelligent chatbots that transform guest communications. Our expertise in conversational AI leverages the integration of both syntax and semantic techniques. In the simplest terms, they relate to language structures and the meaning of words, so you can understand how essential they are for chatbots.
The syntax is an area of study that governs how words are arranged to form coherent sentences. Semantics focus on what those words represent and how they convey information. Incorporating them in our AI-powered chatbot allows us to understand and respond to human questions with unparalleled precision and relevance.
What is syntax analysis?
Syntax analysis (or syntactic analysis) is a methodology for studying conversational language using formal grammar rules. Syntax assigns semantic structure to text. In our AI system, syntax is essential for interpreting and comprehending interactions between hotels and (prospective) guests. By focusing on the structural features of language, we ensure that our algorithms correctly understand user input.
Examples of syntax methodologies:
- Part-of-Speech Tagging (POS Tagging):
- Assigning parts of speech to each word in a sentence (e.g., noun, verb, adjective).
- Parsing:
- Analysing the syntactic structure of a sentence to produce a parse tree showing the grammatical relationships.
What is semantic analysis?
Semantic analysis (or semantics) deals with understanding the meaning and context of language. It is a complicated task for computers since humans usually rely on their intuition and linguistic skills to interpret words, signs and sentence structures. This is why semantic techniques play an important role in our system. By examining the complexities of user queries, we ensure that our answers are accurate and contextually appropriate.
Examples of semantics methodologies:
- Named Entity Recognition (NER):
- Identifying and classifying proper nouns (e.g., names of people, organisations, locations).
- Word Sense Disambiguation (WSD):
- Determining the correct meaning of a word in context when it has multiple meanings.
In order to optimise the accuracy of chatbot answers, the strategies of both fields are incorporated into our AI. By leveraging them, we make sure that the structure and context of the language are known to our models and systems, so they can produce the correct, appropriate answers.
Our proprietary model, developed based on eight years of hospitality-specific data, was built to achieve the highest level of accuracy. This model is exclusive to HiJiffy and has a wide variety of semantic statements that are difficult to understand by other models and which end up being the foundation of our conversational system. This model integrates into our system alongside other machine learning and deep learning algorithms, enhancing its capabilities.
Proximity feature
At its core, the Proximity feature employs syntax and semantics technologies to enhance the capability of the model. It is frequently applied when decoding exceptionally complex user messages with highly complex semantic structures. This model provides the closest answers relevant to the provided topic, so our chatbot would provide correct results.
This model is used in two situations:
- When a guest asks a difficult question (usually containing multiple questions).
- To complement an answer.
Message categorisation model
To improve the accuracy and helpfulness of our virtual concierge (in-stay chatbot), we have developed a Message Categorisation Model which also applies the principles of syntax and semantics. This model categorises incoming messages into three main types: Requests, Complaints, and General Questions (Other Topics).
- Requests: Identifies service requests such as room service, housekeeping, or reservation changes.
- Complaints: Detects issues and problems reported by guests, ensuring prompt attention.
- Other Topics: Handles general questions and informational inquiries.
How it works:
- Guest sends a message to our virtual concierge and depending on the type of message (request or complaint), our system can automatically escalate the conversation to a human agent if necessary.
- If a message is classified as general questions, the chatbot provides immediate and accurate responses based on pre-programmed information.
Let’s take a look at a few examples.
Example 1: Request
- User Input: “Can I get extra towels sent to my room?”
- Categorisation: Request
Example 2: Complaint
- User Input: “The air conditioning in my room isn’t working properly.”
- Categorisation: Complaint
Example 3: General Question (Other Topics)
- User Input: “What time does breakfast start?”
- Categorisation: General Question (Other Topics)
This intelligent categorisation ensures that guest issues are addressed swiftly and efficiently, having a positive impact on the guest experience.
Smart Property Identification
Our innovative Smart Property Identification system is designed to simplify the search for hotels within a specific group. This functionality enables the HiJiffy chatbot to quickly and accurately identify a group’s properties when a potential guest searches by hotel name or enters a specific location.
Let’s see how it works through a specific example.
Example 1 – Hotel name search
- User Input: “Hotel Central Lisboa”
- Smart Property Identification: Hotel Central Lisboa
Example 2 – Location search:
- User Input: “I’m looking for a hotel in New York City.”
- Smart Property Identification: All properties in New York City
Sentences summarisation
The Sentences Summarisation feature handles instances where guests provide lengthy and detailed messages. By summarising these messages, our system can efficiently extract the core intentions and requests, ensuring that the chatbot can respond accurately and promptly (Note: this feature is only available on certain channels, like Instagram).
When a guest sends a long message, the Sentences Summarisation feature analyses the text to identify the essential points. This process involves:
- Text Analysis: The system scans the entire message to identify the main topics and key phrases.
- Core Intent Extraction: It extracts the primary intentions and requests from the text, filtering out irrelevant details.
- Output: The guest is presented with a set of buttons with principal intentions identified in the question.
This summarised input allows the chatbot to understand the guest’s needs more quickly and provide a relevant and accurate response.
Here’s an example to illustrate that in practice.
- User Input: “Hi there! I’m planning to visit your hotel next month with my family, and we are really excited. However, we have a few specific requirements. We need a room with a city view, wheelchair accessibility, and extra bedding for my two children. Can you confirm this and provide the total cost for a week?”
- Summarised Output: The system provides the guest with the following buttons:
- FAQ – Room types and views
- FAQ – Accessibility
- FAQ – Extra bedding
- Book a room
Translations System
One of our systems that makes extensive use of these syntax and semantics techniques is our translation system. Our advanced translation system minimises language barriers, allowing for perfect communication, regardless of the guest’s native language.
If you are interested in learning more about various technologies used in Aplysia, explore a section of our website dedicated to our artificial intelligence, follow HiJiffy on LinkedIn and subscribe to our newsletter in the footer.
More articles in the Explained series
Sources
This article is based on technical contributions by Eduardo Machado from HiJiffy’s AI Team.
